Build a 10x Hermes AI Agent: Complete Setup Guide
Craig Hewitt walks through building a production-grade AI agent using Hermes, covering cloud hosting setup, model configuration, personality tuning, skill creation, multi-channel communication via Telegram, autonomous scheduling with cron jobs, and profile management—transforming a basic chatbot into a powerful autonomous co-worker.
Three Paths to Building AI Agents
Build from Scratch vs. Agent Frameworks
Three main approaches exist: building entirely from scratch (not recommended), using agent-in-a-box solutions like Claude Code or Codex (easy but minimally configurable), or deploying an always-on open-source framework like Hermes (maximum customizability and power for production use).
Why Hermes Over Open Claw
Hermes, created by Nous Research, has become the best agent framework available, surpassing Open Claw (which was acquired by OpenAI). Hermes excels particularly in memory handling and persistent agent state across sessions.
Cloud Infrastructure & Hosting Setup
Why Cloud Over Local Hardware
Running agents on local machines (Mac mini) creates reliability issues—power loss, travel, sleep mode, and configuration crashes make the agent unavailable. Cloud VPS hosting ensures the agent runs 24/7 without physical intervention, though security must be locked down carefully.
Hostinger VPS Setup Process
Select the KVM 2 plan for sufficient memory and storage, choose a 24-month billing cycle for best discount, and use one-click deploy to automatically provision and configure the Hermes agent without manual server setup.
Docker Containerization for Safety
Hermes runs inside a Docker container on the VPS, which isolates the agent in a sandbox. This prevents runaway access or unintended damage while maintaining full functionality and easy terminal access through the Hostinger dashboard.
Model Configuration & Harness Matching
Model-Harness Matching is Critical
A harness is the framework that drives the model; matching the right model to the harness is essential for agent performance. Hermes is optimized for GPT-4.5 via OpenAI's Codex API, which costs $20/month and provides ample tokens for agent operations.
OAuth Setup via Terminal
Connect the model by running 'hermes model' in the terminal, which generates an OAuth code. Paste this code into OpenAI's authentication page to authorize Hermes to use your API plan without exposing credentials directly.
Web Dashboard for Model Management
Hermes comes with a self-hosted web UI that eliminates the need to work in the terminal. This dashboard allows easy model switching, configuration, and agent interaction through a user-friendly interface instead of command-line intimidation.
Personality & Persistent Memory
Soul.md: Persistent Personality File
Soul.md is a configuration file that defines the agent's persistent personality, tone, communication style, directness, warmth, and how much pushback to give. It remains constant across all sessions, unlike temporary personality overlays.
Intersession Memory Compounds Over Time
Unlike Claude Code or Codex, Hermes remembers all previous sessions indefinitely. When you return to the agent, it recalls past conversations and context, eliminating the need to re-explain yourself and allowing the agent to start further toward your goals each time.
Skills: Self-Improving Capabilities
What Are Skills
Skills are reusable, well-crafted prompts saved as callable functions. Instead of repeating detailed instructions every time, you invoke a skill by name (e.g., '/skill social_media_copy') and provide only the topic, dramatically reducing repetition.
Hermes Self-Writes Skills Automatically
Hermes can automatically create new skills when you ask it to or when it detects a repeated pattern. However, this can lead to skill bloat, so manual oversight is necessary. You can also browse trusted skill hubs like skills.sh (Vercel) or provide examples for Hermes to model.
Security Warning: Malware in Skills
Many skill repositories now contain embedded malware and hacker attacks. Rather than installing skills directly, review them first, show them to your agent, and ask it to create a similar skill based on the example—this prevents blind execution of untrusted code.
Creating a Social Media Copy Skill
By providing Hermes with a reference skill (e.g., Corey Haines' social media guidelines), you can ask it to create a custom skill. Hermes invokes its skill-authoring skill, pulls the reference from GitHub, and generates a new callable skill covering LinkedIn, X, Instagram, and TikTok.
Multi-Channel Communication & Gateways
Gateways: Single Entry Point for Communication
A gateway is the communication layer that allows information to flow in and out of the agent. Multiple gateways can be active simultaneously (Telegram, Slack, Discord, web UI, terminal, APIs, custom apps), all routing through a single agent instance.
Telegram Integration via QR Code
Set up Telegram communication by scanning a QR code in the Hermes web dashboard. This creates a bot via BotFather, links it to your user ID, and establishes a secure two-way channel between your Hermes agent in the cloud and Telegram on any device.
Home Channel for Autonomous Notifications
The home channel is where Hermes delivers results from scheduled cron jobs and cross-platform messages. This ensures autonomous tasks (like morning briefings) reach you through your preferred communication channel.
Autonomous Scheduling with Cron Jobs
What Are Cron Jobs
A cron job is a scheduled task that runs at a specific time on a computer. In Hermes, you can create autonomous tasks like 'morning brief' that execute daily at 6 AM, analyze your email and calendar, and deliver a summary to Telegram without manual intervention.
Creating Your First Autonomous Task
Use the command 'cron on profile morning_brief' to schedule a task. Specify the time (e.g., 6:00 AM), the action (e.g., 'take a look at my email and calendar and give me a concise summary'), and the delivery channel (e.g., Telegram). The agent then executes this autonomously every day.
Profiles: Multi-Functional Agent Instances
Profiles as Functional Sub-Agents
Profiles are different instances or personalities of the same agent, each optimized for specific functional areas. For example, a default 'chief of staff' profile can oversee operations, while specialized profiles handle YouTube strategy, social media, email, or business coaching.
Creating a Specialized Profile: Paddy (YouTube Agent)
Create a new profile by naming it (e.g., 'Paddy' after YouTube strategist Paddy Galloway), defining its purpose ('YouTube strategy agent modeled after Paddy Galloway'), and choosing whether to inherit skills, memories, and state from the default profile.
Profile Inheritance and Cloning
New profiles can clone everything (memories, sessions, skills, state) from the default profile or start fresh. Cloning is useful if the agent is mature; starting fresh is better for early-stage agents to avoid confusion about which profile to call.
Avoid Profile Proliferation
Don't create too many profiles early on—it can confuse the main agent about which profile to invoke. Master the default profile and agent framework first, then add specialized profiles once you have strong operational control.
Advanced Connections: MCP & Plugins
Model Context Protocol (MCP) Explained
MCP is a predictable communication layer between AI agents and external services (Google Docs, Notion, Asana, GitHub, etc.) or other AI agents. It ensures the agent knows exactly what capabilities and data it can access from each connected service.
MCP vs. Plugins
MCP is typically an atomic, lightweight protocol for connecting to a single service or API. Plugins are more complete packages bundling multiple related functionalities. Both extend agent capabilities but require careful permission management.
Security Approval for Dangerous Operations
When you ask Hermes to add an MCP or plugin, it performs a security scan and requests approval if the operation involves high-risk interpreters (npm, Python, etc.). This prevents the agent from silently executing potentially dangerous code.
Limit Information Access Early On
Be very cautious about how much information and access you grant your agent early in its lifecycle. Add MCPs and plugins gradually as you gain confidence in the agent's behavior and understand the security implications.
Configuration & Kanban Board
Config.yaml: The Agent's Blueprint
All agent settings live in config.yaml, a text file in the agent's folder accessible via the web dashboard or terminal. If something goes wrong or you need to manually tune the agent, this is where all nitty-gritty configuration happens.
Kanban Board for Task Organization
Hermes includes a built-in Kanban board (similar to Trello) that displays all agent tasks in columns: To Do, Scheduled, Ready, and In Progress. This provides visual context for what the agent is working on and allows you to create, track, and manage tasks from a single interface.
Why Kanban Beats Terminal-Only Workflows
Working purely in the terminal is intimidating and lacks context. The Kanban board provides a visual, organized workspace where you can see task details, create items, and configure the agent—transforming agent interaction from chaotic to professional.
From Zero to Production: The 10 Things Summary
The 10 Essentials for Production AI Agents
Craig covers 10 key elements: (1) choosing Hermes framework, (2) cloud VPS hosting, (3) Docker containerization, (4) model-harness matching (GPT-4.5), (5) web dashboard UI, (6) soul.md personality, (7) intersession memory, (8) skills and self-improvement, (9) multi-channel gateways (Telegram), (10) cron jobs for autonomy, plus profiles and Kanban boards for organization. This transforms a basic chatbot into a production-grade AI co-worker.
Notable quotes
The point of having an agent that's always on is that it's always on. — Craig Hewitt
Once you have this set up, you have literally the most powerful piece of technology maybe ever at your disposal. — Craig Hewitt
Hermes is a self-improving AI agent because it keeps getting smarter for you and it writes its own skills. — Craig Hewitt
Action items
- Sign up for Hostinger VPS using KVM 2 plan and 24-month billing cycle; apply coupon code 'Craig Hewitt' for discount.
- Deploy Hermes via Hostinger's one-click Docker deployment; save generated passwords and API keys in a password manager.
- Connect GPT-4.5 model via OpenAI Codex API by running 'hermes model' in terminal and completing OAuth authentication.
- Customize soul.md file to define your agent's persistent personality, tone, and communication style.
- Create at least one skill (e.g., social media copy) by either browsing trusted skill hubs or asking Hermes to generate one based on a reference example.
- Set up Telegram integration by scanning the QR code in the Hermes dashboard and confirming bot creation via BotFather.
- Create your first cron job (e.g., 'morning_brief') to schedule an autonomous daily task at a specific time.
- Create a specialized profile (e.g., YouTube agent) for a specific functional area and define its purpose in the profile description.
- Test multi-channel communication by sending a message to your Hermes agent via Telegram and verifying the response.
- Review the Kanban board to visualize and manage all agent tasks; create a test task to confirm the workflow.